GreenGrid AI

The problem

• Missouri households carry a disproportionate energy burden: the project cites roughly 20,000 St. Louis City households alone spending 6%–30% of income on utilities, while limited local coordination lets peak loads strain the grid.
• Without access to real smart meters, there was no way to prototype neighborhood-level demand forecasting and personalized guidance without first simulating realistic energy-consumption data.

The approach

• Built as an end-to-end AWS-native pipeline: simulated smart meters and smart plugs publish MQTT messages representing real-time consumption, ingested through AWS IoT Core into S3, processed by Lambda functions for optimization and forecasting (with SageMaker predicting next-day usage), then Bedrock generates personalized, multilingual sustainability tips from the resulting usage patterns stored in DynamoDB.
• Delivered through a responsive AWS Amplify web/mobile app talking to the backend over REST APIs, built and deployed solo, from edge-data simulation through to the live frontend.
• Deliberately scoped as a backend-first proof of concept: prioritized getting the full data pipeline working end-to-end over polishing the UI, and said so directly rather than presenting a rough interface as a finished product.

What broke (and how it got fixed)

• Simulating believable energy-consumption data without access to physical smart meters was a real engineering problem on its own, not a shortcut; every downstream forecast and recommendation depended on getting that simulation right.
• Coordinating data flow cleanly across four separate AWS services (Lambda, S3, DynamoDB, Bedrock) took real troubleshooting, since a break at any handoff silently breaks everything downstream of it.
• Building solo under a hackathon deadline meant explicitly triaging scope: the UI was left basic and not fully polished on purpose, so the available time went into the data pipeline instead.

Outcome

• The full pipeline works end to end and was demoed live, not just described: simulated smart meters → MQTT → AWS IoT Core → S3 → Lambda → Bedrock → an AWS Amplify frontend, deployed and reachable rather than only running locally.
• The generative-AI sustainability tips were explicitly built and tested as a proof of concept, not shipped as a finished feature: the project’s own writeup calls it "just a proof of concept" that "showed real promise," stopping well short of calling it a finished capability.

Stack

Honest scope

A solo hackathon build (AWS Breaking Barriers Virtual Challenge, June 2025) running on simulated smart-meter data, not real household devices; the project’s own roadmap names connecting real smart meters and piloting with actual households as future work, not something already done. The UI was intentionally left basic to prioritize the backend pipeline.